The morning starts with your watch announcing that recovery is at 42%, deep sleep was scarce and the session would be better postponed. The question worth asking before you rebuild the plan: how well does that number actually correspond to what happened overnight?

The gold standard for measuring sleep is polysomnography (PSG): an EEG, eye movements and muscle tone, scored by hand against the criteria of the American Academy of Sleep Medicine. Wearables measure something entirely different — movement and heart rate — and infer the stages from that. Enough comparisons have accumulated for us to be specific.

Six devices put to the test

SLEEP Advances, April 2025. 62 adults spent a night in a sleep lab wearing PSG and between two and four wearables each at the same time.

The devices tested were the Fitbit Charge 5, Fitbit Sense, Withings ScanWatch, Garmin Vívosmart 4, WHOOP 4.0 and Apple Watch Series 8.

The key finding is an asymmetry:

Sensitivity (detecting sleep): above 90% for every device. If you are asleep, the tracker sees it.
Specificity (detecting wakefulness): 29.39–52.15%. If you are lying awake or you woke in the middle of the night, the tracker usually does not see it.

The direct consequence is a systematic overestimation of sleep time:

WHOOP 4.0: +24.46 minutes
Apple Watch Series 8: +19.60 minutes

Most devices differed significantly from PSG on total sleep time, sleep efficiency, wake after sleep onset and light sleep.

For deep sleep the best of the wristbands tested was the WHOOP 4.0, which correctly classified 69.63% of the epochs PSG scored as N3. That is the best result in the set, and it still means that almost a third of deep sleep is scored wrong.

Rings versus wristbands

The largest validation study of a ring deserves its own section. Sleep Medicine, 2024: the Oura Ring Gen3 with the OSSA 2.0 staging algorithm against multi-night ambulatory PSG, 96 participants and 421,045 thirty-second epochs.

sensitivity 94.4–94.5%
specificity 73.0–74.6%
accuracy 91.7–91.8%
staging accuracy: from 75.5% (light sleep) to 90.6% (REM)

Specificity of 73–75% against 29–52% for the wristbands is a marked improvement. The ring takes its signal from a finger, where blood flow and the pulse waveform read more reliably than at the wrist, and that does appear to give it a genuine advantage.

But even here: roughly one epoch of wakefulness in four is counted as sleep.

What follows from this

Total sleep time is the most trustworthy number. It is the closest to the truth on every device, even if it is biased upward by 20–25 minutes.

Sleep stages are the weakest. Light sleep is the hardest to score, and stages are exactly what those attractive “recovery scores” are assembled from.

The error is systematic, not random. That is good news: if your tracker consistently adds 20 minutes, the trend still means something. Comparing yourself with yourself a week ago works. Comparing the absolute number with the advice to “sleep 8 hours” does not.

A tracker understates how bad a bad night was. Precisely because wakefulness is the hardest thing for it to detect: you tossed and turned for an hour and the device credited you with sleep. It will paint the night after a race or a flight as better than it really was.

An important caveat. All these data come from healthy adults, not from athletes in a training block. How far the numbers carry over to someone with a high training load and high heart rate variability is a separate question, and an open one.

How to apply this

Look at total sleep time and the weekly trend, not at the stage breakdown.
Mentally subtract 20–25 minutes from the sleep time shown, if you wear a wristband.
Do not rebuild the plan around a single night. One night is one measurement with a known systematic error.
How you feel beats the score. If recovery reads 90% and you are wrecked, you are right and the device is wrong.
A ring is more accurate than a wristband at detecting wakefulness, if that is the question you are asking. The difference is real but not absolute.
Do not switch devices in the middle of a build — numbers are not comparable between brands, and you will lose your baseline.
If you suspect a genuine sleep disorder — apnoea, persistent insomnia — a tracker will not diagnose it. You need a proper sleep study.

Key points

SLEEP Advances 2025: 62 adults, a night in the lab, six devices against PSG.
Every device detects sleep with sensitivity above 90%; wakefulness is detected with specificity of 29.39–52.15%.
Overestimation of total sleep time: WHOOP 4.0 +24.46 min, Apple Watch S8 +19.60 min.
The best result for deep sleep was WHOOP 4.0 at 69.63% of N3 epochs correctly classified.
Oura Gen3 (Sleep Medicine 2024, 96 people, 421,045 epochs): accuracy 91.7–91.8%, specificity 73.0–74.6%, staging from 75.5% (light) to 90.6% (REM).
The error is systematic — trends are informative, absolute numbers are not.
A tracker makes a bad night look better than it was.
The data come from healthy adults, not from athletes under load.

Sources: «A performance validation of six commercial wrist-worn wearable sleep-tracking devices for sleep stage scoring compared to polysomnography», SLEEP Advances, 2025. https://doi.org/10.1093/sleepadvances/zpaf021 · «Validity and reliability of the Oura Ring Generation 3 (Gen3) with Oura sleep staging algorithm 2.0 (OSSA 2.0) when compared to multi-night ambulatory polysomnography», Sleep Medicine, 2024. https://www.sciencedirect.com/science/article/pii/S1389945724000200